OpenAI employee claims they have already achieved AGI, keeping it under wraps

Just days after the release of its latest model, “O1,” a technical team member from OpenAI stirred the pot by suggesting that the company has already crossed the AGI threshold. He claims OpenAI’s systems are now “better than most humans at most tasks.”

While OpenAI’s achievements continue to push the boundaries of AI capabilities, claims of AGI—however defined—invite scrutiny. Image Credit: Reuters

OpenAI has found itself at the centre of a heated discussion about artificial general intelligence (AGI) following a bold statement by one of its employees. Just days after the release of its latest model, “o1,” Vahid Kazemi, a technical team member, stirred the pot by suggesting that the company has already crossed the AGI threshold—at least by his interpretation.

Kazemi’s claim, made in a post on X (formerly Twitter), hinges on a nuanced definition of AGI. According to him, OpenAI’s systems are now “better than most humans at most tasks,” though not necessarily superior to all humans at every task. This perspective, while provocative, has drawn mixed reactions for its unconventional framing of AGI.

A controversial definition of AGI

Kazemi’s stance redefines what many perceive as AGI. Rather than asserting that the AI surpasses skilled experts in specific domains, he argues that its ability to handle a wide variety of tasks, albeit not flawlessly, positions it beyond human competition in sheer scope. This perspective challenges traditional notions of AGI as human-like intelligence capable of mastering any task to the level of a specialist.

He defended the capabilities of large language models (LLMs) like o1, likening their operation to following a “recipe” akin to the scientific method: observing, hypothesising, and verifying. He suggested that the learning process of LLMs mirrors human intuition, which itself is shaped by trial and error over time.

Timing and business implications

The timing of Kazemi’s remarks is noteworthy. His statements emerged shortly after OpenAI quietly removed “AGI” from the terms of its high-profile deal with Microsoft, leaving industry watchers speculating about the company’s strategic intentions. Some view Kazemi’s claim as a reflection of internal excitement, while others interpret it as an overreach aimed at stirring public interest or justifying OpenAI’s technological trajectory.

Critics argue that even if OpenAI’s models demonstrate remarkable versatility, they remain far from competing with human workers across the board in the labour force. Practical AGI, as many envision it, would entail outperforming humans in a wide range of tasks with consistent reliability, which remains an elusive benchmark.

The bigger picture: Progress or hype?

Kazemi’s post underscores the growing debate about AGI definitions and milestones. While OpenAI’s achievements continue to push the boundaries of AI capabilities, claims of AGI—however defined—invite scrutiny. For now, most experts agree that the world has yet to see an AI capable of competing with humans in generalised, meaningful ways.

If and when such a system emerges, it will undoubtedly demand attention, not just for its technical prowess but for its profound societal and economic implications. Until then, the debate about what constitutes AGI is likely to persist, fuelled by bold statements like Kazemi’s and the rapid advancements in AI technology.